Language Change in Online Social Networks


Li Lei, Limor Raviv & Phillip Alday


Centre for Language Studies, Radboud University
Max Plank Institute for Psycholinguistics
Nijmegen, The Netherlands


RUSE, Manchester, The UK
20/08/2019


People

https://github.com/LangLEvoI/langchangeinnet/blob/master/ruse.slides.html

@Limor_Raviv
@palday

Overview

1.Modelling language change (Fagyal et al., 2010)


2.Visualization in space: Hubs & loners


3.Online social networks

1. Modelling language change

1.1 Background: S-curve

(Source: Fagyal et al. (2010), Lingua)

1.2 In-degree-biased network model

(Source: Fagyal et al. (2010), Lingua)

- The listener ponits to the speaker, e.g. A -> D: L(A), S(D), ind(D)++;

- a higher in-degree, a higher chosen probability, e.g. P(L(A)|S(D)) = 1/6, P(L(A)|S(C)) = 1/3;

- Intuitively, it makes sense in life, e.g. Twitter (followers and following)

1.3 Pseudocode


glob_init()  # random 900 nodes and 7,561 edges, 8 variants
for time in range(40000):  # 40,000 iterations
    for node in range(900):
        selection()  # every node chooses a neighbour via probability
        exchage()  # variant exchange, e.g. var[A] = var[D]

1.4 Result

Multiple s-curves
Double s-curves
S-curve
(Source: Fagyal et al. (2010), Lingua)

1.5 Replication

In [4]:
run(900, 7561, 40000, 'artificial_network')

Interaction %: 100%|##########| 40000/40000 [10:35<00:00, 62.92it/s]

1.6 Manipulations: Indegree-biased selection

a) Random selection (chaos)
b) Fixed selection (chaos)
(Source: Fagyal et al. (2010), Lingua)


- Rule's Plausibility

1.7 Manipulations: Hubs & loners

c) No hubs (chaos)
d) No loners (one s-curve)
(Source: Fagyal et al. (2010), Lingua)
- Hubs: Variant-transmitters;
- Loners: Variant-keepers
Still abstract to understand?

2. Visualization in space

2.1 More clearly and directly to understand

In [47]:
run(100, 400, 500, 'artificial_network')

Interaction %: 100%|##########| 500/500 [00:06<00:00, 72.39it/s]
In [48]:
viz_space(100, 400, 500, 'artificial_network')

Plotting %: 100%|##########| 501/501 [00:01<00:00, 298.75it/s]

3. Online xocial networks

3.1 Epinions

This is a who-trust-whom online social network of a a general consumer review site Epinions.com. Members of the site can decide whether to "trust" each other. All the trust relationships interact and form the Web of Trust which is then combined with review ratings to determine which reviews are shown to the user.

-- (Richardson et al., 2003)


  • 75,879 nodes & 508,837 edges

3.2 Indegree distribution

In [5]:
indegree_figures('artificial_network', 'epinions')

3.3 Result

In [14]:
run(75879, 508837, 40000, 'epinions')

Interaction %: 100%|##########| 40000/40000 [2:41:42<00:00,  4.15it/s]

3.4 Manipulations: Different structure

In [18]:
run(900, 7561, 40000, 'epinions')

Interaction %: 100%|##########| 40000/40000 [08:57<00:00, 37.91it/s]

3.5 Manipulations: More edges

In [20]:
run(900, 27000, 40000, 'epinions')

Interaction %: 100%|##########| 40000/40000 [11:51<00:00, 56.25it/s]

3.6 Conclusion

  • More edges lead to a longer and smoother fixation of linguistic variants;
  • Language change is more difficult in a high-density network (Milroy, 1987)

Future work

  • More "socially realistic" update rules, e.g. weighted edges;
  • Statistical significance of results;
  • More large-scale and representative data, e.g. twitter

References

  • Fagyal, Z., Swarup, S., Escobar, A. M., Gasser, L., & Lakkaraju, K. (2010). Centers and peripheries: Network roles in language change. Lingua, 120(8), 2061-2079.

  • Milroy, L. (1987). Language and social networks. Oxford: Basil Blackwell.

  • Richardson, M., Agrawal, R., & Domingos, P. (2003). Trust management for the semantic web. Proceedings of the 2nd International Semantic Web Conference. 351–368.

Thank you for your attention!

@Limor_Raviv
limor.raviv@mail.huji.ac.il
@palday
phillip.alday@mpi.nl
@LangLEvoI
leili@science.ru.nl